In any field of characteristic 2, every element is its own negative. So on the surface this makes sense, notwithstanding the unintuitive enum corner case behaviour

They should have defined true as -1 (all bits set) instead, as in some other languages like Forth and BASIC.

I wonder if [[=std::bitmask_type]] also enables compiler warnings that a switch statement is not exhaustive if it does not cover all possible flag combinations.

I thought attributes were reserved for compiler-specific semantics like calling conventions. Here they're just being used as stropping for a language feature with syntactic significance!